What a Quantitative Equity Rating Actually Measures

Key Takeaways

Definition: a quantitative equity rating condenses measurable characteristics of a company – valuation, profitability, price momentum, risk – into a single score, computed identically for every stock in a universe. It is a disciplined aggregation of signals, not an opinion.
It is a relative ranking, not a forecast: a high score means "better placed than its peers on the chosen criteria, today." It does not say the stock will go up, or when.
The building blocks come from research: valuation (Fama and French), momentum (Jegadeesh and Titman), profitability (Novy-Marx), quality (Piotroski, Asness and coauthors). These documented premia remain debated – risk compensation or anomaly – and nobody guarantees their persistence.
A rating is judged statistically: return spreads between score deciles, and an information coefficient of around 0.05 for a good signal. A quantitative model's strength is a small edge applied to a large number of decisions, not flashes of brilliance.
Signals erode: across 97 published anomalies, measured returns are 26% lower out-of-sample and 58% lower post-publication (McLean and Pontiff). A rating lives, gets re-estimated, and decays.
The key reflex: faced with a score, the right questions never change – which inputs, which comparison universe, which neutralizations, what out-of-sample validation, what implementation costs.

Introduction

"Quant score: 87/100." "Rated A on 5 criteria." Model-driven stock ratings are everywhere – at data providers, inside banks, across asset managers. The number is crisp, the scale intuitive, the temptation immediate: buy the top-rated names, avoid the rest.

But what exactly does that number measure? The question deserves better than an intuitive answer, because a quantitative rating is neither a price prediction nor a judgment about a company's quality in the everyday sense. It is a precise statistical object, with a construction, documented strengths, and equally well-documented limits.

This article takes the mechanism apart: where quantitative ratings come from, how a score is computed, what it really says, how to tell whether it contains information – and why even the best scores fade over time. It follows on from our article on backtest biases, which described how simulations flatter themselves; this one describes the instrument those simulations are meant to validate.

Where Quantitative Ratings Come From

Equity analysis was long a craft: an analyst covers a few dozen companies, reads their accounts, meets their management, and issues a recommendation. That work has value, but it does not scale – nobody covers the thousands of stocks in a global universe that way, and two analysts rarely apply the same criteria in the same manner.

Quantitative ratings grew out of two findings from academic research. First: certain measurable company characteristics are associated, on average and over long periods, with differences in future returns – the factor literature. Second: a criterion applied mechanically to an entire universe escapes coverage limits and individual inconsistency. Instead of ten deep opinions, the model produces thousands of shallow but systematic assessments – and that breadth is precisely what gives it statistical power, as we will see below.

A quantitative rating is the industrialization of a simple idea: measure the same things, the same way, across all stocks, on the same date, and condense the result into a ranking.

The Building Blocks: Factor Families

What a rating contains depends on the signals it uses. Four families dominate both the literature and practice:

Valuation ("value"): stocks that are cheap relative to their fundamentals – price-to-book, earnings yield – have historically delivered higher average returns than expensive ones. It is one of the central results of Fama and French, a pillar of their 1992 and 2015 models.
Momentum: stocks that outperformed over the past 3 to 12 months tend to keep doing so over the next 3 to 12 months. Jegadeesh and Titman documented spreads on the order of 1% per month between recent winners and losers in 1993, robust across the variants of their protocol.
Profitability and quality: at comparable valuations, companies with durable profitability, solid balance sheets, and conservative accounting have historically performed better. Novy-Marx established it for gross profitability; Piotroski's score aggregates nine accounting signals; Asness, Frazzini, and Pedersen generalized it under the name "quality."
Low risk: low-volatility and low-beta stocks have historically delivered better risk-adjusted returns than classical theory predicts – the anomaly explored by, among others, Frazzini and Pedersen.

Two cautions apply. First, the interpretation of these premia is contested: compensation for real risk, investor behavior, or statistical artifact – the debate is unresolved, and the answer conditions their persistence. Second, the literature has produced hundreds of candidate factors: Harvey, Liu, and Zhu counted 316 in 2016 and concluded that a good share are probably false discoveries, arguing for higher evidentiary thresholds. A serious rating rests on a few robust families rather than a collection of fragile anomalies.

From Measurement to Score: The Cross-Section

Turning raw measurements into a grade follows a standard cross-sectional mechanic: all stocks in a universe are compared with one another on the same date.

Each signal is first standardized – typically as a z-score, the distance from the universe mean expressed in standard deviations – to make heterogeneous quantities comparable. The scores are then combined, often as a weighted average, into a composite that produces the final ranking. A deliberately tiny example, with five fictional companies and three signals:


Company	Earnings yield	ROE	12-month perf.	Composite score	Rank
Echo SA	2% (z = −1.27)	25% (z = 1.37)	+30% (z = 1.57)	0.56	1
Delta SA	10% (z = 1.40)	15% (z = −0.22)	+9% (z = −0.15)	0.34	2
Bravo SA	3% (z = −0.94)	22% (z = 0.89)	+18% (z = 0.59)	0.18	3
Alpha SA	8% (z = 0.73)	12% (z = −0.70)	−5% (z = −1.29)	−0.42	4
Charlie SA	6% (z = 0.07)	8% (z = −1.34)	+2% (z = −0.72)	−0.66	5

Small as it is, the example already shows the essential point: Echo SA ranks first even though it is the most expensive stock in the universe, because its profitability and momentum outweigh its valuation under the chosen weights. Delta SA, the valuation champion, comes second. A composite score is a quantified trade-off between qualities that do not travel together – and what "top-rated" means depends entirely on which signals are used and how they are weighted.

In practice, refinements are added that change the outcome considerably: sector neutralization (compare a bank with banks, not with software companies, or the rating becomes a permanent sector bet), outlier treatment, handling of missing data, weights that may be estimated rather than fixed. Two ratings built on the same factor families can therefore disagree markedly on a given stock – there is no "objective" score, only scores whose construction is more or less transparent.

What a Rating Really Says – and What It Does Not

Three properties define the exact scope of a quantitative score.

It is a relative ranking. A z-score is computed against a universe: the same stock can be "cheap" among Swiss equities and "expensive" among European ones. A high score says nothing in absolute terms; it says "better placed than the others, on these criteria, in this universe, on this date."

It is a probabilistic statement, not a single-stock forecast. The factor literature is about portfolio averages: highly rated stocks have, on average, historically outperformed poorly rated ones. On any individual stock, the score is often wrong – a hit rate barely above 50% is everyday life for the best models. Quantitative discipline means accepting that individual fallibility and offsetting it with numbers: many positions and many rebalancings, so the small statistical edge has room to materialize.

It is a snapshot with a defined horizon. Every signal has its own shelf life: momentum goes stale in weeks, valuation moves in quarters. A rating is dated and calibrated for a horizon; using it at another horizon means using a different instrument.

The practical consequence deserves stating plainly: a quantitative rating is neither a buy recommendation nor a verdict on a company. It is one component of a process – universe screening, portfolio construction, risk control – and its value depends on the process around it.

How a Rating Is Judged: Deciles, Information Coefficient, Breadth

A score can look impressive and carry no information. Three standard tools help tell the difference.

Decile spreads: sort the universe by score, cut it into ten groups, and compare their subsequent returns. An informative signal produces a positive spread between the top and bottom deciles, but above all a roughly monotonic progression from one decile to the next: a spread driven by two lucky extremes, with no order in the middle, is a classic symptom of fragility.
The information coefficient (IC): the correlation between scores on one date and the returns realized afterward. The orders of magnitude surprise newcomers: in the practitioner literature descending from Grinold and Kahn, an IC of 0.05 already characterizes an exploitable signal. Stock by stock, the information content of a score is small.
Breadth: Grinold's fundamental law of active management formalizes the intuition that saves everything: a process's information ratio grows with the IC multiplied by the square root of the number of independent decisions. A tiny edge, applied with discipline across thousands of decisions, becomes statistically significant; the same edge across ten positions remains noise.

All three measures are computed on historical data – and therefore inherit every trap described in our article on backtests: survivorship bias, data ahead of its publication date, overfitting, ignored costs. A decile spread gross of costs, measured on today's universe projected into the past, can be nothing but an artifact. A rating's validation is worth exactly what its protocol is worth.

Why Ratings Lose Their Shine

Three forces work against any score that is published or exploited.

Post-publication erosion. McLean and Pontiff re-examined 97 anomalies documented by academic research: the associated returns are on average 26% lower outside the original sample and 58% lower after academic publication. The gap between those two numbers suggests investors read the research and arbitrage part of the anomaly away; the first number suggests another part never existed outside the sample. Either way, a signal's forward-looking expectation is smaller than its history.

Crowding. When large pools of capital follow the same scores, highly rated stocks get bid up, compressing the premium – and creating a new risk: simultaneous unwinds, where the crowded exit of related strategies amplifies declines precisely in the top-rated names.

Regimes. No factor family performs in every environment: value has endured multi-year deserts, momentum has suffered brutal crashes during market rebounds. A composite score smooths these cycles without removing them. The question is not "is the factor working right now?" but "will the process survive its worst stretch?" – a matter of portfolio construction and governance as much as of signal.

Do Not Confuse: Quant Ratings, Credit Ratings, ESG Scores

The word "rating" covers unrelated objects. A credit rating (Moody's, S&P, Fitch) assesses the probability that a borrower honors its debt: it concerns the bond, not the attractiveness of the stock. An ESG score measures environmental, social, and governance characteristics, under methodologies that diverge sharply across providers. A quantitative equity rating ranks stocks on characteristics that research associates with future returns. All three can coexist on the same company with opposite conclusions – a firm can be an excellent credit risk, a mediocre ESG scorer, and a top-ranked quant name. None of them "corrects" the others: they do not measure the same thing.

Conclusion

A quantitative equity rating measures one precise thing: a stock's relative position, within a given universe, on characteristics that research associates on average with return differences. Its strength is not clairvoyance – stock by stock, it is often wrong – but discipline and breadth: the same criteria, applied everywhere, all the time, with a small statistical edge that only repetition makes exploitable.

It is also a fallible, dated, and perishable object: dependent on its inputs, its universe, and its weights, validated on a past that flatters, eroded by its own popularity. The questions that separate a disciplined score from a decorative number fit on one line: which signals, which universe, which neutralizations, what out-of-sample validation, what costs and capacity. A rating provider that answers those five questions precisely takes its instrument seriously; a score that does not answer them remains just a number.